Estimating QCD background in single muon channel Alexander D

Oldenzaal April '10 2 introduction motivation: qcd has large theoretical uncertainty and needs to be understood from data. we use the isolated-muon-side of the story… goal: give an estimate of the number of qcd events after all top cuts to be used in analysis. input for example template method (see Hegoi’s talk) d0 significance ABCD method fit method problems solution: the matrix method data: distributions we want to use conclusion & plans

Oldenzaal April '10 3 d0 significance the transverse impact parameter: how far in (x,y) from (0,0) use the fact that extra muons are mainly non-prompt from b’s d0 significance = d0/sigma(d0) normalized to unity prompt = muon in ttbar(mu), matched, extra = muon in qcd note that d0 is actually d0-corrected, i.e. with respect to the primary vertex: d0 corr =

Oldenzaal April '10 4 ABCD method [1] w+jets, ttbar, single top, qcd: AB CD C = AD/B C = 11.7 ± 2.6 MC = 12.7 ± pb -1 general idea: if 2 variables are uncorrelated then A/B = C/D and A/C = B/D assumption: signal-like muons in C, rest only qcd-muons

Oldenzaal April '10 5 ABCD method [2] systematic error due to boundaries: change values and see how prediction changes: two features: jumps due to limited qcd events with high weights overall increase due to more ‘signal’-like muons in D C = 11.7 ± 2.6(stat.) (syst.) MC = 12.7 ± pb -1 Note: the method estimates qcd in region with d0 significance < 3. This is NOT what the analysis uses. How to solve this?

Oldenzaal April '10 6 fit method top cuts: all top, w+jets, single top, qcd under 0.1: ttbar(mu): ± 1.1 other top: 29.5 ± 0.5 qcd: 34.6 ± 6.2 predicted: 45.3 ± 6.1 (stat.) ± 4.5 (syst.) use etcone20/p T distribution (relative isolation) assumption: signal-like muons are isolated, qcd-muons are not 9.6pb -1

Oldenzaal April '10 7 problems… Fit method: great results, but only without overlap removal  statistics too low… new default isolation cut also ptcone30. how to implement in fit? ABCD method: results in region not quite top default: the d0 significance cut is not used also the extra isolation cut is hard to implement the solution: matrix method no need for continuous cuts: two isolation cuts is no problem easy to choose regions with high statistics

Oldenzaal April '10 8 matrix method the method relies on the distinction between loose and tight muons where idea is that we have isolated muons (ttbar, W+jets, Z+jets) and non-isolated muons (qcd) tight is a subset of loose tight is usually the final selection used in the analysis loose is usually kinematically the same as tight, but less or no quality requirements N L = N L R + N L F N T = N T R + N T F = ε R N L R + ε F N L F where ε is the efficiency for fake/real lepton to move from L(oose) to T(ight). this then yields:

Oldenzaal April '10 9 loose - tight Loose: MuId muons p T > 20 GeV |η| < 2.5 isTight (MCP flag) Tight 1: + etcone20/p T < ptcone30 < 4 GeV + dR to jet > 0.3 (if jet-p T > 0.5 muon-p T ) Tight 2: + d0 significance < 3.0 ε the definitions should aim for: tight is dominated by signal-like muons loose is mainly qcd muons we will use the d0 significance as an extra cut  good cross check

Oldenzaal April '10 10 measure what? where? N L is just the number of loose muons in your selected top-sample same for N T ε R has to come from a signal-muon dominated sample: Z  µµ same as for muons from top? same as in ttbar-environment? ε F will be measured using a QCD enriched sample: low MET is the efficiency flat with respect to MET? what about contamination from W’s, Z’s? the matrix method also has problems, but they all seem well documented and solvable…

Oldenzaal April '10 11 distribution in data [1] run MinBias selected lumiblocks: 117 – 207 BCID == 1  1,165,191 events same ‘loose’ as before except p T > 2 GeV find 1072 muons ε = 64%

Oldenzaal April '10 12 distribution in data [1] run MinBias selected lumiblocks: 117 – 207 BCID == 1  1,165,191 events same ‘loose’ as before except p T > 2 GeV find 1072 muons ε = 64%

Oldenzaal April '10 13 distribution in data [2] ‘our’ variable: the d0 significance first calculate the d0 corrected then calculate d0/sigma(d0)

Oldenzaal April '10 14 conclusions & outlook both fit and ABCD method yield good results with nice systematic errors both however have problems, either statistics (overlap removal) or intrinsic (since they don’t use the exact right cuts) solution is there: matrix method our main distributions look really good so far in data. maybe we can trust MC after all… need to get the method working convince the rest of top-analysts that d0-significance is a great idea the matrix method has quite some “haken en ogen” and “voeten in de aarde”, but there seem to be no show stoppers!

back-up slides

Oldenzaal April '10 16 run

Oldenzaal April '10 17 ABCD method 2 Jet events, no etmiss cut  high statistics, but correlated: C = ± 44.7  qcd: ± 39.1 using just the qcd numbers: ± 44.4 signal in C: 31.1 ± 0.5 other top in C: 32.2 ± Jet events, top cuts  low statistics, but uncorrelated: C = 11.5 ± 2.6  qcd: 12.7 ± 2.6 using just the qcd numbers: 6.4 ± 2.5 signal in C: ± 0.9 other top in C: 25.4 ± 0.4 C = A*D/B method works well in top-region, but difference in 2 jet events is large. might be correlation, but “using just qcd numbers” also not spot on in top-region errors are only statistical note that for scaled histogram the error is not just sqrt(N) but scaled with scaling factor. 9.6pb -1

Oldenzaal April '10 18 ABCD method [numbers] 2 Jet events, no etmiss cut  high statistics, but correlated: A:: ± events (qcd: ± ) B:: ± events (qcd: ± 128.4) C:: ± 40.9 events (qcd: ± 39.1) D:: ± 31.4 events (qcd: ± 31.3) AD/B = ± 44.7 using just qcd: ± 44.4 signal in C: 31.1 ± 0.5 other top in C: 32.2 ± 0.5 other bckg in C: ± Jet events, top cuts  low statistics, but uncorrelated: A:: ± 36.7 events (qcd: ± 36.7) B:: ± 37.1 events (qcd: ± 37.0) C:: ± 4.2 events (qcd: 12.7 ± 3.6) D:: 11.9 ± 2.6 events (qcd: 6.5 ± 2.6) AD/B = 11.7 ± 2.6 using just qcd: 6.4 ± 2.5 signal in C: ± 1.1 other top in C: 25.4 ± 0.5 other bck in C: ± pb -1

Oldenzaal April '10 19 fit method details…. the fit is Landau*pol3 (7 parameters, p(i)) extrapolation is done using a straight line  from f(0.1) to 0. error is calculated using V(f) = GV(x)G -1 where V(x) is the covariance matrix and G is df/dp(i) varying a given parameter with its error gives fit as shown important is f(0.1) for top-cuts region: f(0.1) = 7.2 ± 1.0 Integral: 45.3 ± 6.1 df/dp(5) 9.6pb -1

Oldenzaal April '10 20 Fit method [overlap removal] top-cuts, OR(jet-muon), no etmiss-cut top-cuts, OR(jet-muon) under 0.1: qcd: 75.7± 9.3 (pred: 52.8 ± 23.5) signal: under 0.1: qcd: 19.6 signal: fit breaks down at the moment with overlap removal: no statistics above pb -1